Introduction

As the trend of miniaturization continues, the next generation of electronic devices will likely be downsized to molecular scale. Since the first molecular device, the so-called Avram-Ratner (AR) rectifier1, numerous molecular electronic devices with unusual properties such as high-performance transistors2,3,4, light driven switches5,6,7 and novel spintronics devices8,9,10 have been suggested experimentally and/or theoretically. Previous studies have clearly shown that many exciting features of these devices are closely related to the intrinsic electronic properties of the molecular junction. Understanding the electronic properties of the molecular junction under the operating conditions now has become the central issue in the theoretical modeling of these devices.

Due to highly reduced dimensions, the properties of molecular devices are often strongly size-dependent and very sensitive to their chemical environment. A reliable and efficient parameter-free first-principles modeling technique is therefore highly desired. There are two major difficulties in atomistic first-principles modeling of these devices. Firstly, the device is connected with two electrodes so that the whole system is not periodic and the so-called scattering boundary conditions (SBCs) have to be considered. Secondly, since the molecular devices always operate under a finite bias, the bias-induced nonequilibrium effects need to be correctly taken into account, which presents a great challenge to the modeling. The first difficulty has been resolved. It has been shown that the mean-field equation from density functional theory (DFT) with SBCs can be solved either by plane-wave based methods11 or by non-equilibrium Green’s functions’ (NEGF) techniques12. The so-called DFT + NEGF method also included part of nonequilibrium effects and was then implemented with various computational schemes13,14,15 and widely applied in studying all kinds of nonequilibrium molecular scale devices. Over the last decade, the DFT + NEGF method has achieved great success in terms of providing insights into experiments and inspiring novel devices. While, the aforementioned second difficulty still remains because DFT is a theory for ground-state (GS) properties. The applicability of DFT for nonequilibrium systems is questionable. In fact, several recent works16,17 have argued that when a system is under a finite bias, the basis of GS-DFT, the first Hohenberg-Kohn theorem (HKT)18 may not be applicable any more.

In our previous work, without rigorous proof, we suggested that additional degrees of freedom (could be an additional density) besides the total electron density are needed to determine the properties of a nonequilibrium quantum system16. In this paper, starting from the Hershfield’s nonequilibrium quantum statistics19, we derive a density functional theory for the steady-state (SS) properties of molecular junctions under a finite bias. We prove that with given two densities, the total electron density and the density of current-carrying electrons, the Hamiltonian of an effective equilibrium system (that can be mapped to the desired steady-state nonequilibrum junction) is uniquely determined and in turn the steady-state properties of the corresponding nonequilibrium system are also determined, confirming that GS-DFT is in principle incorrect for nonequilibrium systems. A self-consistent mean-field approach based on the two densities is then derived by minimizing the energy of the effective equilibrium system. The theory is called SS-DFT (comparing to GS-DFT) in this paper. We apply the theory to a carbon-nanotube (CNT)/Benzene junction. Compared with the standard GS-DFT based DFT + NEGF method, the SS-DFT gives significantly lower energies at all biases and importantly, we show that for a limiting case (which will be shown later) for which the electric current is negligible, the SS-DFT predicts the correct electronic properties of the molecular center while the DFT + NEGF does not.

Model and Theory

We illustrate in Fig. 1 the model for a molecular junction under a finite bias. The model consists of two reservoirs that are in their own equilibrium (the source and drain) and a molecular scale device region. We assume mean-field Hamiltonians in two reservoirs so that two chemical potentials, μL and μR, can be defined for the source and drain, respectively. Electrons in the device region, however, are interacting. Without losing generality, we assume μL ≥ μR. The bias voltage is defined as eVb = μL − μR. The overall Hamiltonian for such an open system can be written as consisting of the Hamiltonian for the non-interacting left (right) reservoir , the interacting device region and the tunneling term .

Figure 1
figure 1

Model of a molecular junction under a finite bias.

The model consists of source, drain and a molecular scale device region. The source and drain are assumed to be non-interacting and in local equilibrium.

When the nonequilibrium system is in steady state, Hershfield19 showed that the nonequilibrium ensemble average of any physical observable can be calculated using an equilibrium-like statistics,

where accounts for nonequilibrium distribution and can be written as following20,

in which the creation operator creates an electron incident from the left (right) reservoir. Defining two number operators and , , the distribution operator becomes . Hershfield’s theory provides a very attractive way for investigating steady-state properties of nonequilibrium systems without the need of complicated time-dependent perturbation. However, due to the difficulty of working out the distribution operator from a many-body Hamiltonian, the application of the theory has been rather limited. The purpose of this paper is to derive a DFT-like first-principle method based on Hershfield’s nonequilibrium statistics that can be applied to realistic systems.

We introduce another set of number operators, and , defined as and . Eq. 1 can then be rewritten as,

where and . Eq. 3 clearly shows that the steady-state properties of the nonequilibrium system can be obtained from a corresponding effective equilibrium system with the Hamiltonian and the chemical potential .

Apparently, and have physical meanings of the total number of electrons and the number of current-carrying electrons, respectively. After transforming the number operators to real space by Fourier transform, two electron densities, ρt(r) and ρn(r), can be defined (Eq. 4). The integration of these densities in real space yield and , respectively.

We are now ready to derive the first HKT-like theorem for the nonequilibrium steady state of the system. We will prove that the two densities, ρt(r) and ρn(r), uniquely determine the Hamiltonian of the effective equilibrium system and in turn the steady-state properties of the nonequilibrium system (Eq. 3). The proof is given in the following discussions. In general, the Hamiltonian reads as follows,

where the kinetic and e-e interaction terms are universal for all systems. Unlike the ground-state case where the Hamiltonian is completely determined by the external potential , the effective-equilibrium Hamiltonian here is determined by three terms, the external potential , the bias voltage Vb and the nonequilibrium number operator . The bias voltage Vb is a global term that is associated with boundary conditions provided by two reservoirs and cannot be determined by total electron density alone as discussed in our previous work16 so that the first HKT for GS-DFT is not applicable here. In order to prove that two densities determine everything including Vb, we first assume that there are two different effective equilibrium Hamiltonians, and , which have two ground states and respectively and bear the same set of densities ρt and ρn. Then we have,

In above derivations, the temperature is set to zero (T → 0), so the equilibrium ensemble average becomes a ground-state integral. and are ground-sate energies of and , respectively. is the distribution function defined in Eq. 3. We can also start from the ensemble average of and similarly,

Combining Eq. 6 and Eq. 7, we get

which cannot be true. Thus ρt and ρn determine up to a trivial additive constant and in turn, determine the ensemble average of any physical observables (Eq. 3) of the nonequilibrium steady state. Note that the ground state of the effective equilibrium is assumed to be non-degenerate in the above derivations.

The next step is to derive a mean-field approach to calculate the two electron densities. For this purpose, we write the ground-state energy of the effective equilibrium as a functional, , where ρe ≡ ρt − ρn. By generalizing the local density approximation (LDA) to nonequilibrium cases, the energy functional becomes,

In Eq. 9, T is the kinetic energy, VH is the Hartree potential that should be calculated under proper open-system boundary conditions and exc is the exchange correlation energy density of nonequilibrium uniform electron gas.

To make the calculations feasible, we assume that the electron densities in two mean-field equilibrium reservoirs are not affected by interactions in the device region due to the screening in contacts, which is consistent with the conventional transport modeling13,15. As a consequence, the electron densities in two reservoirs can be treated as boundary conditions and the variations of the densities can be done only in the device region. Before applying the variational principle to Eq. 9, following the spirit of the Kohn-Sham approach of GS-DFT, we introduce a non-interacting reference system of which the single-electron states bear real equilibrium and current-carrying electron densities in the device region,

In the equations above, single-electron state is normalized in the device region; fe and fn together specify the nonequilibrium distribution

where fFD is the Fermi-Dirac distribution function at zero temperature. When an infinitesimal variation is applied to any state ϕ (either or ) in the reference system,

we have the variational condition .

In order to make the variational principle sensible, we add a constraint that the total number of electrons in the device region is conserved. There are also two boundary conditions arising from the assumption that the variations are only done in the device region, which read (∂ΩD denotes the boundaries of the device region)

To solve the constrained variational problem, we use the method of Lagrange multipliers. Define

where are real and are the occupation numbers for equilibrium (current-carrying) electrons. Let

the variational condition then becomes

After some straightforward derivations, we arrive at the following two mean-field equations, one for current-carrying electrons and another one for equilibrium electrons that do not contribute to current,

where the two effective potentials are

As shown in Eq. 18, two effective potentials differ in the third term, the so-called exchange-correlation potential Vxc. The effective potential for current-carrying electrons has an additional term which is proportional to bias voltage. Note that Eq. 18 can be easily generalized to include generalized gradient approximation (GGA) (See supporting information). The two mean-field equations (Eq. 17) are coupled with each other through the exchange-correlation functional that depends on both ρe and ρn. These equations have to be solved self-consistently (see the self-consistent procedure in Figure S1 of supporting information). By assuming that the correlation functional is the same as GS case and using the nonequilibrium exchange functional we previously derived16,17, the SS-DFT can be applied to real molecular junctions.

Results and Discussions

We have implemented the SS-DFT into the widely used computational package SIESTA21. The theory is then applied to a CNT-benzene-CNT junction that can be fabricated in lab with radical chemistry. Energetics and I-V characteristics from SS-DFT are compared to those obtained from the conventional DFT + NEGF approach via the function TranSIESTA15. As shown in the inset of Fig. 2a, the CNT-benzene-CNT molecular junction consists of two semi-infinite metallic (5, 5) CNTs and a benzene molecule in between. In order to form good contacts, two H atoms in the benzene molecule are taken away (which can be done in chemical synthesis). Both CNTs are closed at one end with a C30 cap. The atomic structures of CNTs, the device region and the distance between two leads are optimized with full quantum calculations. In all calculations, we used single-ζ basis set and PBE exchange-correlation functional22 corrected by the nonequilibrium exchange energy16,17. In the Brillouin zone, 1 × 1 × 30 k-point sampling for lead calculations is employed. The energy and force (for structure relaxation) convergence criteria are set to 10−4eV and 0.04 eV/Å, respectively.

Figure 2
figure 2

(a) Current-Voltage characteristics of CNT-Benzene-CNT junction from SS-DFT and DFT + NEGF calculations. Inset: Atomic structure of the junction. (b) Energy difference between SS-DFT and DFT + NEGF methods defined as δE = ESS−DFT − EDFT+NEGF. Both δE and are presented. The SS-DFT always gives lower energies when the bias voltage is not zero.

Two methods, SS-DFT and DFT + NEGF, yield similar I-V curves for the junction under study except that between bias voltages from 0.05 and 0.2 V, SS-DFT predicts significantly lower currents. In Fig. 2b, we plot the energy difference between two methods. Note that E and are ensemble averages of and , respectively. Clearly, SS-DFT gives lower energies in all bias voltages, which is expected since compared with DFT + NEGF, SS-DFT searches the lowest-energy steady state in a much bigger Hilbert space. For equilibrium situation when ρn = 0, SS-DFT reduces to GS-DFT. Consequently, there is no essential difference between SS-DFT and DFT + NEGF at zero bias. Both methods predict a prominent negative-differential-resistance (NDR) peak around 0.05 V where the SS-DFT peak occurs at a slightly smaller bias than the DFT + NEGF one. This can be understood by the fact that the two methods generate different electronic properties of the device region and in this case, the lowest unoccupied orbital (LUMO) of the benzene molecule calculated from SS-DFT exits the bias window earlier than the one computed from DFT + NEGF (see details in Figure S2 of supporting information), leading to different positions of NDR peaks.

We now consider a limiting case of which the electronic properties of a certain part are known and check whether or not the SS-DFT is able to reproduce those known properties. The limiting case is shown in Fig. 3a where the distance between the center molecule and the leads is set to more than 6 Å. In this case, the electric current is essentially zero and the center molecule is decoupled from the two leads. Although the whole system is driven out of equilibrium by the external bias (0.5 V of voltage for this case), the center molecule is in local equilibrium, which means that its properties can be determined by GS-DFT. We therefore perform calculations for the electronic properties of the center molecule using three different methods, the GS-DFT for the junction under zero bias, the SS-DFT and DFT + NEGF for the junction under 0.5 V of bias voltage. In Fig. 3, we show the calculated projected density of states (PDOS) of the center molecule around the Fermi Energy. The SS-DFT agrees with GS-DFT (DFT in the figure) very well, while the DFT + NEGF generates different electronic structures. To further illustrate the difference between different methods, in Fig. 3c, we plot the difference between total electron densities around the center molecule calculated from different methods, δρ = ρDFT − ρDFT+NEGF in left and δρ = ρSS−DFT − ρDFT+NEGF in right. Clearly, DFT + NEGF yields quite different total electron densities from other two methods and the densities from DFT and SS-DFT are similar. Note that in GS-DFT calculations, there is no electric field across the molecule, which leads to unsymmetrical δρ in the left panel of Fig. 3c. We also performed a simple GS-DFT calculations with an external electric field (without the presence of two leads) and then the δρ around the molecule between GS-DFT and DFT + NEGF becomes symmetrical (see Figure S3 in supporting information).

Figure 3
figure 3

(a) The atomic structure of a limiting case where the distance between the center molecule and each of two leads is more than 6 Å. In this case, the center molecule is essentially isolated. (b) Projected density of states (PDOS) of the center molecule calculated from three different methods, DFT for zero bias case, DFT + NEGF and SS-DFT for the junction under a bias voltage of 0.5 V. EF in the figure represents . (c) The difference in the total electron densities calculated from different methods, δρ = ρDFT − ρDFT+NEGF and δρ = ρSS−DFT − ρDFT+NEGF. The isosurface value is 5 × 10−5 Bohr−3 in both plots where the red (blue) color denotes positive (negative) value.

Conclusion

In summary, we present in this paper a density functional theory for the steady state of nonequilibrium molecular junctions. We prove that two electron densities, ρt and ρn, uniquely determine the steady-state properties of nonequilibrium quantum systems. Two mean-field equations that are coupled with each other, one for equilibrium electrons and another one for current-carrying electrons, are derived. The theory is applied to a CNT-benzene-CNT junction. Calculations based on the SS-DFT are compared with those from the conventional DFT + NEGF calculations via TranSIESTA. It is found that the SS-DFT always leads to energetically more stable steady states and lower electric currents than conventional GS-DFT based transport calculations. We also show that for a limiting case in which the center molecule is essentially isolated, the SS-DFT is able to predict correct local equilibrium properties. The theory lays a solid ground for atomistic first-principles modeling of steady-state nonequilibrium quantum systems and paves the way for computational understanding and design of complex molecular scale devices.

Additional Information

How to cite this article: Liu, S. et al. Density Functional Theory for Steady-State Nonequilibrium Molecular Junctions. Sci. Rep. 5, 15386; doi: 10.1038/srep15386 (2015).